Anthropic has issued a striking warning to the world. AI systems are rapidly approaching the ability to design and train their own successors, and humanity may not be ready for what comes next.
In a new report titled 'When AI Builds Itself,' the San Francisco-based company reveals that AI is already accelerating the development of AI systems, presenting data showing Anthropic engineers today ship 8x as much code per quarter as they did between 2021 and 2025, largely because Claude is writing most of it.
As of May 2026, more than 80 percent of the code merged into Anthropic's codebase was authored by Claude. The company's most advanced model, Claude Mythos Preview, can now complete tasks that would take a skilled human up to 12 hours, up from just four minutes two years ago. anthropicanthropic
The pace is accelerating sharply. The length of tasks that AI models can reliably complete on their own has been doubling roughly every four months. If the trend holds, Anthropic projects that by 2027, AI systems could handle work that takes a person weeks.
The company describes a near-future scenario, 'closing the loop', where agents become capable enough to build and train models themselves, so future versions of Claude could be continuously improved by Claude itself.
That prospect carries serious risks. Full recursive self-improvement might increase the risks of humans losing control over AI systems. Anthropic warns that if AI can fully build its own successors, the ways we secure, monitor, and shape its behaviour all grow much more important.
In the most alarming scenario outlined, rare occurrences of misalignment present in today's models could compound as the models build their successors, growing more frequent but less understood until we lose control of them.
Anthropic acknowledges it cannot unilaterally halt progress, noting that a unilateral pause by one lab would change who the front-runner is, but would not create the wider deliberative process that is currently missing. The company is calling for verified, multilateral coordination, and says the window to act is now.
